Investigate with evidence. Close with a disposition that sticks.

The case, not the chat

An investigation in In-Sight is a case with an owner, a timeline, and a scope. Related findings and alarms are rolled into it, the affected tags are charted across the window, operator round readings from the period are attached, and hypotheses are ranked by the evidence for and against each. The team works the case in one place, and the case remains queryable after it is closed.

• Launch from any finding, roll related findings and alarms into the case
• Timeline of the affected tags with the finding, alarm, and round markers overlaid
• Ranked hypotheses with the evidence behind each
• Disposition that writes back to every rolled-up finding
• Recurrence tracking after closure, so a fix that did not hold is surfaced

A copilot that has read the plant's documents

Datasheets, SOPs, P&ID notes, and manuals are parsed, including drawings via vision, chunked, embedded, and retrieved with hybrid search. When the copilot proposes that a thickener underflow excursion is consistent with flocculant make-up strength, it cites the page it drew that from. When it cannot source a claim, it abstains rather than guess. It drafts the RCA narrative, ranks likely causes by evidence, generates charts from a plain-language question, and answers questions across live data, findings, and documents in one place.

The same document intelligence mines engineering limits from those pages and proposes them to an engineer with the citation. The engineer confirms; only then is the limit armed.

Ask In-Sight

Outside a case, any user can ask a question in plain language: which mill had the most findings last month, what the average flotation air was during the excursion, what the SOP says the minimum froth depth should be. Answers cite the data and the page, and charts are generated from the question. Text-to-SQL runs against the historian with the plant's tag labels, so the answer speaks the plant's language.

Institutional memory that survives crew changes

Every investigation, disposition, and root cause stays queryable. When a new finding resembles a closed case, the precedent is shown alongside it with what was done and whether it held. The hard-won knowledge that usually lives in two heads and a shared drive becomes an asset the plant keeps.

Questions we are asked

Does the copilot need internet access?

It needs access to a language model endpoint. That can be a hosted model or an approved endpoint chosen by the site; the platform itself deploys inside the plant network or in a dedicated cloud environment.

How does In-Sight stop the AI from making things up?

Answers are grounded in retrieved passages from the plant's own documents and in the platform's own data, and every claim carries a citation. When retrieval finds nothing to support a claim, the copilot says so and abstains.

What document types can be ingested?

PDF datasheets, SOPs, manuals, batch records, and reports, including scanned pages and drawings, which are processed with vision. Each document's pages are retrievable with page-level provenance.

Can an investigation be started manually, without a finding?

Yes. A finding can be captured manually from an observation or a round reading, and an investigation can be launched from it in the same way.

Related